The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The reasoning metrics dashboard functionality processes command execution history, creating a potential surface for indirect prompt injection via manipulated metadata.
Ingestion points: Untrusted data enters the agent context through command execution logs stored in ~/.claude/.metrics/reasoning/ (referenced in references/metrics.md).
Boundary markers: Absent. The instructions do not specify delimiters or warnings for the agent to ignore potentially embedded instructions within the metadata logs.
Capability inventory: The skill utilizes Read for log access, Write for exporting reports, and Bash for generating visualizations using plotting tools (referenced in references/metrics.md).
Sanitization: Absent. There are no explicit instructions for the agent to sanitize or validate command names or other metadata before processing them for reports or visualizations.
[COMMAND_EXECUTION]: The skill coordinates with shell tools for analytics and visualization purposes.
Evidence: The Tool Coordination section in references/metrics.md explicitly lists Bash as a tool for generating visualizations with plotting tools and Grep for pattern analysis in usage logs.

reasoning-controls